DOJ Endorses Fair Use for AI Training, Reshaping Copyright Economics
Executive Summary
The US Department of Justice filed an amicus brief backing OpenAI and Microsoft in the New York Times copyright case, formally endorsing fair use as federal policy for AI training data. This materially weakens the legal leverage of every publisher and content licensor currently in AI negotiations. Any licensing agreement in Q4 renegotiation should be reopened now — counterparties have lost their primary legal argument, and waiting for a court ruling cedes the leverage window.
The Signal
The US Department of Justice filed an amicus brief on September 2, 2026, siding with OpenAI and Microsoft in the consolidated New York Times copyright lawsuit, arguing that training large language models on copyrighted material constitutes fair use. The DOJ drew a legal distinction between the act of copying during training — which it contends never makes works publicly available — and model outputs, which it says rarely bear substantial similarity to source material. The brief directly challenged the US Copyright Office's earlier report rejecting blanket fair use for AI training, dismissing former Register Shira Perlmutter's conclusions as carrying no binding legal authority. Perlmutter was fired by the Trump administration shortly after publishing that report. The case, filed in Manhattan federal court in late 2023, is the most consequential pending copyright action in the AI industry.
What Changed
For the first time, the executive branch of the US government has formally endorsed fair use as the operative legal framework for AI training at commercial scale. That endorsement converts what was a contested legal theory into a stated position of federal policy. AI developers can now build and expand training pipelines on copyrighted corpora with the US government as an explicit legal ally — meaningfully reducing the litigation risk calculus that had been constraining training data decisions and licensing negotiations.
Why It Matters
The DOJ brief doesn't just help OpenAI win a case — it restructures the entire economics of AI model development in favor of companies that can train at scale without licensing costs. That is the commercial earthquake beneath the legal argument. Training data licensing, which had been quietly becoming a significant line item for AI developers and the enterprise brands building on their infrastructure, just lost most of its negotiating leverage. Publishers, news organizations, and content owners who had been extracting licensing fees — or holding out for them — are now negotiating against a federal government position that their core ask is legally unnecessary. The New York Times's damages theory, worth billions on paper, looks considerably weaker when the executive branch of the plaintiff's own government is filing briefs against it. For the brands and agencies building proprietary models on internal corpora, this matters less directly. But for those relying on foundation model providers — which is most of the market — the downstream effect is real: training pipelines get cheaper, model iterations accelerate, and providers face less friction expanding into new content domains. That translates into faster capability releases and lower API costs over the 12-to-24-month horizon. What becomes genuinely obsolete is the licensing-as-moat strategy. Any content business that built a forward-looking revenue model around mandatory AI licensing fees is now sitting on a structurally weaker asset. The strategic logic driving this is straightforward: the Trump administration has decided that US AI competitive dominance outweighs domestic content industry protection, and it is using the DOJ as the instrument to make that trade explicit. That priority is not going to reverse before the Manhattan court rules.
Marketing Impact
creative
Generative creative teams can now build and expand training pipelines on licensed or unlicensed corpora with materially lower legal exposure. Brand voice models, style-tuned image generators, and vertical-specific copy tools that stalled in legal review are now viable builds. The constraint shifts from 'is this permitted' to 'is this useful.'
martech
Martech vendors building proprietary AI layers — personalization engines, predictive content tools, semantic search — face reduced friction when expanding training data sourcing. Licensing cost lines that were being baked into enterprise contracts as compliance buffers can be renegotiated or removed, compressing vendor costs and accelerating product roadmaps.
marketing ops
Internal AI governance reviews that had been blocking training pipeline expansion on cost and legal grounds lose a significant portion of their rationale. Marketing ops teams running or procuring AI tooling can accelerate procurement cycles and retire the 'training data provenance' risk register item that legal had been flagging since 2024.
The Exploit
Opportunity
Enterprise brands and agencies building proprietary models or fine-tuning foundation models on third-party content corpora can now accelerate those programs without provisioning licensing reserves. Legal budget previously held against training-data liability can be redeployed. More concretely: any content licensing renegotiation scheduled for Q4 2026 should be reopened now — counterparties have lost their primary legal leverage.
Risk
The district court could still rule against fair use, and an appellate process running to 2028 or beyond would restore licensing risk. Moving without legal sign-off exposes brands to secondary infringement claims if outputs are challenged.
The Move
General Counsel and CMO jointly review any content licensing agreement currently in negotiation or renewal before October 2026, using the DOJ brief as explicit leverage to reduce fee commitments or exit restrictive exclusivity terms — with legal sign-off as the gate and renegotiated contract terms as the success checkpoint.
First-Mover Advantage
Gains
Teams that restart paused training programs before the Manhattan ruling lock in a 12-to-18-month model-maturity lead over competitors still waiting for legal certainty that, at the federal policy level, has effectively arrived.
Risks
The district court could still rule against fair use, and an appellate process running to 2028 or beyond would restore licensing risk. Moving without legal sign-off exposes brands to secondary infringement claims if outputs are challenged.
Window
The advantage window stays open until a federal court ruling contradicts the DOJ position — the signal to watch is the Manhattan district court's summary judgment decision, expected in early 2027.
Winners & Losers
Winners↑
Foundation model providers training on web-scale corpora
The DOJ brief converts fair use from a litigation gamble into a stated federal policy position, materially reducing the legal risk premium baked into training pipeline decisions at OpenAI, Anthropic, Google DeepMind, and Meta Platforms. Licensing cost pressure eases, training data scope can expand, and model iteration cycles accelerate — all of which flows downstream into faster capability releases and more competitive API pricing. These providers should use the legal tailwind to aggressively expand training data coverage into verticals — legal, medical, financial publishing — where licensing friction had previously constrained them.
Enterprise marketing teams building on foundation model APIs
Cheaper, faster model iteration at the provider layer translates directly into lower API costs and richer capabilities for the enterprise teams running personalisation engines, content generation pipelines, and AI-assisted media buying on top of those models. The 12-to-24-month horizon for cost compression is now more credible, making the business case for scaling AI-dependent marketing operations easier to defend internally. Teams should lock in multi-year API agreements now, before the capability step-changes that a less constrained training environment will produce.
AI-native martech and adtech vendors with API-dependent product stacks
Vendors whose products are built on foundation model outputs — generative creative tools, AI copy platforms, predictive audience modellers — benefit from the same upstream cost compression and stand to see their own unit economics improve as provider pricing adjusts. The DOJ position also reduces the governance ambiguity that had slowed enterprise procurement of AI-dependent SaaS, removing a common legal objection from risk and compliance reviews. These vendors should be updating their security and legal FAQs to cite the DOJ position explicitly, shortening sales cycles where copyright exposure was a procurement blocker.
In-house legal and AI governance teams at large advertisers
The DOJ brief gives in-house counsel a federal government anchor for the position that using foundation models does not create downstream copyright exposure from training data — resolving one of the most persistent legal uncertainties slowing AI adoption in regulated categories. Governance teams that had been drafting conservative acceptable-use policies around this risk can now recalibrate toward the narrower, output-similarity standard the DOJ endorses. The practical move is to update AI use policies to reflect the training-versus-output distinction and clear the backlog of AI tool deployments that had been stuck in legal review.
Losers↓
Publisher and news organisation IP licensing desks
Any content business that had built forward revenue projections around mandatory AI training licensing — whether through direct deals like those struck with Associated Press and Axel Springer, or holdout strategies aimed at extracting larger settlements — is now negotiating against an explicit federal government position that their core legal ask is unnecessary. The New York Times's multi-billion-dollar damages theory weakens materially when the executive branch files on the opposing side, and other rights holders watching that case will re-price their leverage accordingly. The defensive play is to shift negotiating focus from training fees to output distribution rights and attribution — the narrower ground the DOJ brief does not foreclose.
Content and data licensing intermediaries positioned as AI training suppliers
The intermediary businesses that had positioned themselves as compliant pipelines between rights holders and AI developers — licensing aggregators, data brokers specialising in clean training sets — lose their primary value proposition if fair use becomes the settled legal framework and developers can train without licensing at all. Deal flow dries up, existing contracts face renegotiation pressure, and the pipeline model itself becomes structurally redundant in the segments most affected. Survival depends on pivoting toward provenance, metadata, and quality-curation services that add value independent of legal necessity.
Strategic Outlook
The Manhattan court is not bound by the DOJ brief, but federal courts rarely rule against an explicit executive branch policy position when national competitiveness is the stated stakes. Expect OpenAI and Microsoft to cite the brief aggressively in summary judgment motions; a ruling before end of Q2 2027 is plausible if the court accelerates on the back of the government's clarity. If fair use is affirmed at the district level, the licensing negotiation landscape collapses almost immediately — holdout publishers lose their best legal leverage before any appeals court can restore it. The parallel effect on model costs is real but lagged: cheaper training pipelines translate into API pricing pressure over a 12-to-18-month horizon rather than overnight. Content businesses that built licensing revenue projections into 2027 budgets should treat those figures as impaired now, not after the ruling. The window for extracting meaningful licensing fees from the major foundation model providers is effectively closing.